@dickpy/dsh-imagegen 1.2.3 → 1.4.0

package/src/model-catalog.ts

@@ -0,0 +1,115 @@
+/**
+ * Model catalog: protocol-family detection and capability annotation shared by
+ * the engine, the settings card, and the panel. Known id patterns map onto the
+ * OpenAI-compatible image protocol families this plugin shapes requests for;
+ * anything unrecognized falls back to the generic OpenAI protocol (best
+ * effort) and is flagged as unknown so the UI can warn without blocking.
+ *
+ * Framework-free (pure data + regex), safe for the client bundle to inline.
+ */
+
+export type ModelFamily = 'gpt-image' | 'dall-e' | 'grok' | 'nanobanana' | 'seedream' | 'zhipu' | 'unknown'
+
+/** Capability/identity annotation for one model id. */
+export interface ModelCatalogEntry {
+  /** Family the request is shaped for. */
+  family: ModelFamily
+  /** Short badge label (en). */
+  label: string
+  /** Short badge label (zh). */
+  labelZh: string
+  /** Whether the family is known (false = best-effort OpenAI protocol). */
+  known: boolean
+  /** Whether the family supports image-to-image edits. */
+  supportsEdit: boolean
+  /** Whether the family natively takes aspect-ratio dials. */
+  supportsAspectRatio: boolean
+  /** Quality tiers the family interprets natively. */
+  qualityTiers: string[]
+}
+
+const ENTRIES: Record<Exclude<ModelFamily, 'unknown'>, Omit<ModelCatalogEntry, 'family'>> = {
+  'gpt-image': {
+    label: 'gpt-image',
+    labelZh: 'GPT 图像',
+    known: true,
+    supportsEdit: true,
+    supportsAspectRatio: false,
+    qualityTiers: ['1K', '2K', '4K'],
+  },
+  'dall-e': {
+    label: 'DALL·E',
+    labelZh: 'DALL·E',
+    known: true,
+    supportsEdit: true,
+    supportsAspectRatio: false,
+    qualityTiers: ['auto'],
+  },
+  grok: {
+    label: 'grok',
+    labelZh: 'Grok',
+    known: true,
+    supportsEdit: true,
+    supportsAspectRatio: true,
+    qualityTiers: ['1K', '2K'],
+  },
+  nanobanana: {
+    label: 'nanobanana',
+    labelZh: 'Nano Banana',
+    known: true,
+    supportsEdit: true,
+    supportsAspectRatio: true,
+    qualityTiers: ['1K', '2K', '4K'],
+  },
+  seedream: {
+    label: 'seedream',
+    labelZh: 'Seedream',
+    known: true,
+    supportsEdit: true,
+    supportsAspectRatio: true,
+    qualityTiers: ['1K', '2K'],
+  },
+  zhipu: {
+    label: 'GLM-Image',
+    labelZh: '智谱图像',
+    known: true,
+    supportsEdit: false,
+    supportsAspectRatio: false,
+    qualityTiers: ['HD'],
+  },
+}
+
+/** Official Gemini image ids served by Nano Banana gateways. */
+const NANOBANANA_GEMINI_IDS = new Set([
+  'gemini-3-pro-image',
+  'gemini-3-pro-image-preview',
+  'gemini-3.1-flash-image',
+  'gemini-3.1-flash-image-preview',
+  'gemini-3.1-flash-lite-image',
+  'gemini-2.5-flash-image',
+])
+
+/** Classify one upstream model id into its request-shaping family. */
+export function describeModel(model: string): ModelCatalogEntry {
+  const id = model.trim()
+  if (/^gpt-image/i.test(id)) return { family: 'gpt-image', ...ENTRIES['gpt-image'] }
+  if (/^dall-e/i.test(id)) return { family: 'dall-e', ...ENTRIES['dall-e'] }
+  if (/^grok-imagine(?:-|$)/.test(id)) return { family: 'grok', ...ENTRIES.grok }
+  if (/^nanobanana/i.test(id) || NANOBANANA_GEMINI_IDS.has(id)) return { family: 'nanobanana', ...ENTRIES.nanobanana }
+  if (/^(?:doubao-)?seedream/i.test(id)) return { family: 'seedream', ...ENTRIES.seedream }
+  if (/^(?:glm-image|cogview(?:-|$))/i.test(id)) return { family: 'zhipu', ...ENTRIES.zhipu }
+  return { family: 'unknown', label: 'unknown', labelZh: '未知协议', known: false, supportsEdit: true, supportsAspectRatio: false, qualityTiers: [] }
+}
+
+/** Conservative fallback for providers whose /models response only has ids.
+ *  Metadata-aware filtering lives in prompt-enhancer.ts; this catches common
+ *  image model naming conventions without treating every unknown model as an
+ *  image model. */
+export function isLikelyImageModelId(model: string): boolean {
+  return /(?:^|[-_.])(?:image|img|diffusion|flux|cogview|imagen|seedream|nanobanana|grok-imagine|dall-e|stable-diffusion|sdxl|pixart|kolors|ideogram|midjourney|recraft|hunyuan|jimeng|wanx|hidream|playground)(?:$|[-_.])/i.test(model.trim())
+}
+
+/** The family a model id routes its request through. */
+export function modelFamily(model: string): ModelFamily {
+  return describeModel(model).family
+}

package/src/presets.ts

@@ -0,0 +1,71 @@
+/**
+ * Built-in provider catalog (presets). A preset is an official or well-known
+ * OpenAI-compatible endpoint with its known model list, so the user only fills
+ * in the API key. The list ships with the package and is served to the
+ * settings card through a host route, so it can later be refreshed online like
+ * the template library.
+ *
+ * Framework-free (pure data), safe for the host routes to serve directly.
+ */
+
+import type { ModelMapping } from './protocol.ts'
+
+/** One built-in provider the settings card can instantiate a channel from. */
+export interface PresetProvider {
+  /** Stable preset id stored on channels created from it ('' = custom). */
+  id: string
+  /** Display name shown in the picker (also the channel's default name). */
+  name: string
+  /** Official base URL prefilled into the channel. */
+  apiUrl: string
+  /** One-line description shown in the picker. */
+  hint: string
+  /** Known model list prefilled into the channel's model catalog. */
+  models: ModelMapping[]
+}
+
+export const IMAGE_PRESETS: PresetProvider[] = [
+  {
+    id: 'volc-ark-seedream',
+    name: '字节 · 火山方舟（Seedream）',
+    apiUrl: 'https://ark.cn-beijing.volces.com/api/v3',
+    hint: '字节跳动官方 Seedream 文生图/图生图入口',
+    models: [
+      { alias: 'seedream-5.0-pro', id: 'seedream-5.0-pro' },
+      { alias: 'seedream-5.0', id: 'seedream-5.0' },
+      { alias: 'seedream-4.0', id: 'seedream-4.0' },
+    ],
+  },
+  {
+    id: 'openai-official',
+    name: 'OpenAI 官方',
+    apiUrl: 'https://api.openai.com/v1',
+    hint: 'OpenAI 官方图像生成接口',
+    models: [
+      { alias: 'gpt-image-2', id: 'gpt-image-2' },
+    ],
+  },
+  {
+    id: 'zhipu-official',
+    name: '智谱 AI 官方',
+    apiUrl: 'https://open.bigmodel.cn/api/paas/v4',
+    hint: '智谱官方 GLM-Image 图像生成接口',
+    models: [
+      { alias: 'glm-image', id: 'glm-image' },
+    ],
+  },
+  {
+    id: 'xai-grok',
+    name: 'xAI（Grok）',
+    apiUrl: 'https://api.x.ai/v1',
+    hint: 'xAI 官方接口：Grok Imagine 系列',
+    models: [
+      { alias: 'grok-imagine-image', id: 'grok-imagine-image' },
+    ],
+  },
+]
+
+/** Look up one built-in provider by id. */
+export function presetById(id: string): PresetProvider | undefined {
+  return IMAGE_PRESETS.find(preset => preset.id === id)
+}

package/src/prompt-enhancer.ts

@@ -1,79 +1,137 @@
-/** OpenAI-compatible chat helpers used by the optional prompt-enhancement UI. */
-
-export interface PromptModelConfig {
-  apiUrl: string
-  apiKey: string
-  model: string
-}
-
-/** Credentials shared by OpenAI-compatible `/models` discovery. */
-export interface ModelListConfig {
-  apiUrl: string
-  apiKey: string
-}
-
-function endpoint(base: string, suffix: string): string {
-  return `${base.replace(/\/+$/, '')}${suffix}`
-}
-
-function headers(apiKey: string): HeadersInit {
-  return {
-    'content-type': 'application/json',
-    ...apiKey.trim() === '' ? {} : { authorization: `Bearer ${apiKey.trim()}` },
-  }
-}
-
-async function responseJson(response: Response): Promise<Record<string, unknown>> {
-  const body: unknown = await response.json().catch(() => undefined)
-  if (!response.ok || body === undefined || body === null || typeof body !== 'object') {
-    const message = body !== null && typeof body === 'object' && typeof (body as { error?: { message?: unknown } }).error?.message === 'string'
-      ? (body as { error: { message: string } }).error.message
-      : `HTTP ${response.status}`
-    throw new Error(message)
-  }
-  return body as Record<string, unknown>
-}
-
-/** List candidates exposed by an OpenAI-compatible endpoint. */
-export async function listOpenAIModels(config: ModelListConfig): Promise<string[]> {
-  if (config.apiUrl.trim() === '') throw new Error('API URL is required')
-  const response = await fetch(endpoint(config.apiUrl, '/models'), { headers: headers(config.apiKey) })
-  const body = await responseJson(response)
-  const data = Array.isArray(body.data) ? body.data : []
-  return [...new Set(data
-    .flatMap(item => item !== null && typeof item === 'object' && typeof (item as { id?: unknown }).id === 'string' ? [(item as { id: string }).id.trim()] : [])
-    .filter(Boolean))]
-    .sort((a, b) => a.localeCompare(b))
-}
-
-/** List chat models exposed by an OpenAI-compatible endpoint. */
-export async function listPromptModels(config: PromptModelConfig): Promise<string[]> {
-  return listOpenAIModels(config)
-}
-
-/** Expand a concise image request into a production-ready image prompt. */
-export async function enhancePrompt(config: PromptModelConfig, prompt: string): Promise<string> {
-  if (config.apiUrl.trim() === '' || config.model.trim() === '') throw new Error('prompt enhancement model is not configured')
-  const response = await fetch(endpoint(config.apiUrl, '/chat/completions'), {
-    method: 'POST',
-    headers: headers(config.apiKey),
-    body: JSON.stringify({
-      model: config.model.trim(),
-      temperature: 0.7,
-      messages: [
-        {
-          role: 'system',
-          content: 'You are an expert image-prompt editor. Expand the user request into one vivid, specific image-generation prompt. Preserve intent and language. Add only useful visual detail: subject, composition, lighting, materials, color, camera/style and quality. Return only the finished prompt, with no preface or markdown.',
-        },
-        { role: 'user', content: prompt },
-      ],
-    }),
-  })
-  const body = await responseJson(response)
-  const choices = Array.isArray(body.choices) ? body.choices : []
-  const content = choices[0] !== null && typeof choices[0] === 'object'
-    ? (choices[0] as { message?: { content?: unknown } }).message?.content
-    : undefined
-  if (typeof content !== 'string' || content.trim() === '') throw new Error('chat model returned an empty prompt')
-  return content.trim()
-}
+/** OpenAI-compatible chat helpers used by the optional prompt-enhancement UI. */
+
+import { isLikelyImageModelId } from './model-catalog.ts'
+
+export interface PromptModelConfig {
+  apiUrl: string
+  apiKey: string
+  model: string
+}
+
+/** Credentials shared by OpenAI-compatible `/models` discovery. */
+export interface ModelListConfig {
+  apiUrl: string
+  apiKey: string
+}
+
+function endpoint(base: string, suffix: string): string {
+  return `${base.replace(/\/+$/, '')}${suffix}`
+}
+
+function headers(apiKey: string): HeadersInit {
+  return {
+    'content-type': 'application/json',
+    ...apiKey.trim() === '' ? {} : { authorization: `Bearer ${apiKey.trim()}` },
+  }
+}
+
+async function responseJson(response: Response): Promise<Record<string, unknown>> {
+  const body: unknown = await response.json().catch(() => undefined)
+  if (!response.ok || body === undefined || body === null || typeof body !== 'object') {
+    const message = body !== null && typeof body === 'object' && typeof (body as { error?: { message?: unknown } }).error?.message === 'string'
+      ? (body as { error: { message: string } }).error.message
+      : `HTTP ${response.status}`
+    throw new Error(message)
+  }
+  return body as Record<string, unknown>
+}
+
+type ModelRecord = Record<string, unknown> & { id: string }
+
+async function listModelRecords(config: ModelListConfig): Promise<ModelRecord[]> {
+  if (config.apiUrl.trim() === '') throw new Error('API URL is required')
+  const response = await fetch(endpoint(config.apiUrl, '/models'), { headers: headers(config.apiKey) })
+  const body = await responseJson(response)
+  const data = Array.isArray(body.data) ? body.data : []
+  return data.flatMap(item => {
+    if (item === null || typeof item !== 'object' || typeof (item as { id?: unknown }).id !== 'string') return []
+    const id = (item as { id: string }).id.trim()
+    return id === '' ? [] : [{ ...(item as Record<string, unknown>), id }]
+  })
+}
+
+function textOf(value: unknown): string[] {
+  if (typeof value === 'string') return [value]
+  if (!Array.isArray(value)) return []
+  return value.filter((item): item is string => typeof item === 'string')
+}
+
+function hasImageGenerationCapability(record: ModelRecord): boolean | undefined {
+  const capability = record.capabilities
+  if (capability !== null && typeof capability === 'object') {
+    const values = capability as Record<string, unknown>
+    for (const key of ['image_generation', 'imageGeneration', 'text_to_image', 'textToImage', 'image_gen']) {
+      if (typeof values[key] === 'boolean') return values[key]
+    }
+    const serialized = JSON.stringify(values).toLowerCase()
+    if (/image[ _-]?generation|text[ _-]?to[ _-]?image/.test(serialized)) return true
+  }
+
+  const taskText = [
+    ...textOf(record.task),
+    ...textOf(record.task_type),
+    ...textOf(record.taskType),
+    ...textOf(record.type),
+    ...textOf(record.model_type),
+    ...textOf(record.modelType),
+    ...textOf(record.tasks),
+    ...textOf(record.description),
+  ].join(' ').toLowerCase()
+  if (/image[ _-]?generation|text[ _-]?to[ _-]?image|image[ _-]?gen/.test(taskText)) return true
+  if (/^image(?:[ _-]?generation)?$/.test(taskText.trim())) return true
+  if (/embedding|rerank|moderation|transcri|speech|audio|video|chat[ _-]?completion/.test(taskText)) return false
+
+  for (const key of ['output_modalities', 'outputModalities', 'supported_output_modalities']) {
+    const modalities = textOf(record[key]).map(value => value.toLowerCase())
+    if (modalities.length > 0) return modalities.includes('image')
+  }
+  return undefined
+}
+
+function isImageModelRecord(record: ModelRecord): boolean {
+  return hasImageGenerationCapability(record) ?? isLikelyImageModelId(record.id)
+}
+
+/** List candidates exposed by an OpenAI-compatible endpoint. */
+export async function listOpenAIModels(config: ModelListConfig): Promise<string[]> {
+  return [...new Set((await listModelRecords(config)).map(record => record.id))]
+    .sort((a, b) => a.localeCompare(b))
+}
+
+/** List only models that advertise or conventionally represent image generation. */
+export async function listImageModels(config: ModelListConfig): Promise<string[]> {
+  return [...new Set((await listModelRecords(config)).filter(isImageModelRecord).map(record => record.id))]
+    .sort((a, b) => a.localeCompare(b))
+}
+
+/** List chat models exposed by an OpenAI-compatible endpoint. */
+export async function listPromptModels(config: PromptModelConfig): Promise<string[]> {
+  return listOpenAIModels(config)
+}
+
+/** Expand a concise image request into a production-ready image prompt. */
+export async function enhancePrompt(config: PromptModelConfig, prompt: string): Promise<string> {
+  if (config.apiUrl.trim() === '' || config.model.trim() === '') throw new Error('prompt enhancement model is not configured')
+  const response = await fetch(endpoint(config.apiUrl, '/chat/completions'), {
+    method: 'POST',
+    headers: headers(config.apiKey),
+    body: JSON.stringify({
+      model: config.model.trim(),
+      temperature: 0.7,
+      messages: [
+        {
+          role: 'system',
+          content: 'You are an expert image-prompt editor. Expand the user request into one vivid, specific image-generation prompt. Preserve intent and language. Add only useful visual detail: subject, composition, lighting, materials, color, camera/style and quality. Return only the finished prompt, with no preface or markdown.',
+        },
+        { role: 'user', content: prompt },
+      ],
+    }),
+  })
+  const body = await responseJson(response)
+  const choices = Array.isArray(body.choices) ? body.choices : []
+  const content = choices[0] !== null && typeof choices[0] === 'object'
+    ? (choices[0] as { message?: { content?: unknown } }).message?.content
+    : undefined
+  if (typeof content !== 'string' || content.trim() === '') throw new Error('chat model returned an empty prompt')
+  return content.trim()
+}